FLUX 3 Max: fal and Black Forest Labs push image-to-video to 1,139 Elo — in 11 seconds flat

• Who: fal, with Black Forest Labs — post-trained by fal Research on the FLUX 3 foundation model.
• What: FLUX 3 Max, a video generation model (text-to-video, image-to-video, keyframes-to-video) built via specialized post-training, reinforcement learning, and inference optimization.
• When: Announced and available October 19, 2026; 75% off for the first three weeks.
• Quality: 1,139 Elo on the Artificial Analysis Video Arena (4,868 votes), +50 points over FLUX 3's 1,089.
• Speed: 10-second video in 10.95 seconds — 22x faster than Seedance 2.5 (277.80 s) at the same reported Elo; 11.5x faster than FLUX 3 (126.44 s).
• Strategy: Extends the post-training playbook behind fal's earlier H3 Max; fal says this is the first iteration and post-training continues.

A frontier model, post-trained rather than rebuilt

FLUX 3 Max is not a new foundation model. According to the announcement, fal Research took Black Forest Labs' FLUX 3 foundation model and ran it through specialized post-training, reinforcement learning, and inference optimization — improving visual fidelity, motion coherence, prompt adherence, and generation performance without training a model from scratch. The pitch is that frontier capability increasingly comes from the system around the model, not just the base weights.

"Model performance doesn't stop improving when base model training ends," said Kaan Oktay, Head of Research at fal. "Through post-training and inference optimization, we're able to improve both the quality of generated video and the efficiency of the underlying model. With FLUX 3 Max, we've achieved a 50-point Elo improvement while making inference more than 11 times faster. And we're continuing to post-train the model to push performance even further."

The approach extends the research behind H3 Max, fal's previously released frontier video generation model, which the company says produced video in 3.49 seconds in its own benchmark while keeping top-tier image-to-video quality. FLUX 3 Max applies the same playbook — specialized post-training plus systems-level inference optimization — to someone else's foundation model.

The numbers: 1,139 Elo, 10.95 seconds, 22x the speed of Seedance 2.5

In preliminary independent human preference evaluations on the Artificial Analysis Video Arena, FLUX 3 Max scored 1,139 Elo across 4,868 votes — placing it among the highest-quality image-to-video models evaluated, with overlapping confidence intervals among the top ranks, per fal's announcement.

The speed claims are where fal is loudest. In the company's latency benchmarks, FLUX 3 Max generates a 10-second video in 10.95 seconds. Seedance 2.5, which fal says achieves the same reported image-to-video Elo, takes 277.80 seconds for the same task — making FLUX 3 Max more than 22x faster at that quality tier.

Against its own foundation, the comparison is direct: original FLUX 3 scored 1,089 Elo, so Max is a 50-point quality gain; original generation latency was 126.44 seconds, so Max is 11.5x faster. That combination — better and dramatically faster — is fal's case that the speed-quality frontier for video diffusion transformers still has room to move.

Unite.AI, covering the launch, noted the model became available the same day it was announced through fal's inference platform — a same-day ship that keeps the Elo claims honest in a way a future release would not.

Why Black Forest Labs opened the door

Black Forest Labs built FLUX 3 as a foundation for others to build on — and FLUX 3 Max is the worked example. "We built FLUX 3 as a foundation for others to build on," said Robin Rombach, co-founder and CEO of Black Forest Labs. "FLUX 3 Max is a great example of what that looks like. fal took our model and customized it for the developers and enterprises on their platform. That's the kind of collaboration FLUX was designed for, and we're excited to see it reach more of the people building with it."

fal CEO Burkay Gur framed the partnership as a strategy, not a one-off: "We believe some of the most important advances in generative AI will come from optimizing models and the systems that run them together. Black Forest Labs has built an exceptional foundation model, and our research team has developed deep expertise in pushing the boundaries of quality and inference performance. FLUX 3 Max demonstrates what's possible when those capabilities come together: frontier video quality at speeds that fundamentally change how developers can build with generative AI."

Available now — at 75% off for three weeks

FLUX 3 Max is available beginning October 19, 2026, through fal's inference platform, with a unified API covering text-to-video, image-to-video, and keyframes-to-video generation. To mark the launch, fal is offering it at 75% off for the first three weeks.

The launch marks the first iteration of the model — fal says it is continuing to post-train FLUX 3 Max, with further improvements to visual quality, generation speed, and overall performance underway. For developers, the message is that today's benchmarks are a floor: the post-training loop is still running. Full details are at fal.ai.